ComfyUI Workflow: LoRA Settings Tester

This ComfyUI workflow is designed to test LoRA (Low-Rank Adaptation) settings and visualize the impact of various parameters on the generated outputs. The workflow includes a plot feature for comparing results side by side, providing an intuitive way to evaluate changes.

Custom Nodes

• Anything Everywhere? / https://github.com/chrisgoringe/cg-use-everywhere

• Bookmark (rgthree) / https://github.com/rgthree/rgthree-comfy

• JWInteger / https://github.com/jamesWalker55/comfyui-various

• LorasForFluxParams+ / https://github.com/cubiq/ComfyUI_essentials

• PlotParameters+ / https://github.com/cubiq/ComfyUI_essentials

• ProjectFilePathNode / https://github.com/MushroomFleet/DJZ-Nodes

• SaveText|pysssss / https://github.com/pythongosssss/ComfyUI-Custom-Scripts

• Searge_LLM_Node / https://github.com/SeargeDP/ComfyUI_Searge_LLM

• Seed Everywhere / https://github.com/chrisgoringe/cg-use-everywhere

• Simple String / https://github.com/chrisgoringe/cg-use-everywhere

• Text to Conditioning / https://github.com/WASasquatch/was-node-suite-comfyui

• UnetLoaderGGUF / https://github.com/city96/ComfyUI-GGUF

• easy showAnything / https://github.com/yolain/ComfyUI-Easy-Use

• FluxSamplerParams+ / https://github.com/cubiq/ComfyUI_essentials

Workflow Features

1. Exposed Parameters

This workflow exposes several key settings for fine-tuning and experimentation:

Model Strength

• Description: Controls the intensity of the LoRA model's influence during image generation.

• Exposed Values: 0.5, 0.6, 0.7, etc.

• Effect: Higher values increase the LoRA model's impact, while lower values reduce it.

Sampler

• Description: Determines the algorithm used for sampling during the diffusion process.

• Exposed Options: ipndm, euler, etc.

• Effect: Different samplers produce varying levels of detail, sharpness, and adherence to prompts.

Scheduler

• Description: Controls the schedule of noise reduction during the diffusion process.

• Exposed Options: beta, simple, etc.

• Effect: Influences the pace and smoothness of the denoising process.

Clip Attention Multiply

• Description: Scales the attention weights for specific tokens in the text prompt.

• Effect: Fine-tunes how strongly individual parts of the prompt affect the output. Values >1 amplify token importance, while values <1 reduce it.

2. Visual Output

The workflow generates a plot showing side-by-side comparisons of the outputs under different parameter settings. This allows for easy evaluation of:

• Variations in model strength.

• Differences between samplers.

• Impact of scheduler choices.

3. Save Image

• Outputs are automatically saved with filenames reflecting the parameter settings used, making it easy to track and compare results.

How to Use

1. Set LoRA Parameters:

   • Adjust the model strength to experiment with the influence of the LoRA model.

   • Choose a sampler and scheduler to explore their effects.

2. Customize Prompt:

   • Enter your desired text prompt in the prompt node.

3. Run the Workflow:

   • Execute the workflow to generate outputs with different parameter settings.

4. Review Results:

   • Use the plot to compare outputs side by side.

   • Saved images can be reviewed in the designated output folder.

Practical Tips

• Start Simple: Use default values and adjust one parameter at a time for clearer insights.

• Experiment with Combinations: Test different combinations of samplers and schedulers for optimal results.

• Use Clip Attention Multiply Sparingly: Small adjustments (e.g., 1.1 or 0.9) can significantly affect results.

---

This workflow provides a robust tool for testing and fine-tuning LoRA settings in ComfyUI, enabling users to achieve their desired artistic or practical outcomes with ease.